Explainable machine learning for diffraction patterns.

Serial crystallography experiments at X-ray free-electron laser facilities produce massive amounts of data but only a fraction of these data are useful for downstream analysis. Thus, it is essential to differentiate between acceptable and unacceptable data, generally known as 'hit' and 'miss', respectively. Image classification methods from artificial intelligence, or more specifically convolutional neural networks (CNNs), classify the data into hit and miss categories in order to achieve data reduction. The quantitative performance established in previous work indicates that CNNs successfully classify serial crystallography data into desired categories [Ke, Brewster, Yu, Ushizima, Yang & Sauter (2018). J. Synchrotron Rad.25, 655-670], but no qualitative evidence on the internal workings of these networks has been provided. For example, there are no visualization methods that highlight the features contributing to a specific prediction while classifying data in serial crystallography experiments. Therefore, existing deep learning methods, including CNNs classifying serial crystallography data, are like a 'black box'. To this end, presented here is a qualitative study to unpack the internal workings of CNNs with the aim of visualizing information in the fundamental blocks of a standard network with serial crystallography data. The region(s) or part(s) of an image that mostly contribute to a hit or miss prediction are visualized.

Enhancing Renal Tumor Detection: Leveraging Artificial Neural Networks in Computed Tomography Analysis.

BACKGROUND Renal cell carcinoma is one of the most common cancers in Europe, with a total incidence rate of 18.4 cases per 100 000 population. There is currently significant overdiagnosis (11% to 30.9%) at times of planned surgery based on radiological studies. The purpose of this study was to create an artificial neural network (ANN) solution based on computed tomography (CT) images as an additional tool to improve the differentiation of malignant and benign renal tumors and to aid active surveillance. MATERIAL AND METHODS A retrospective study based on CT images was conducted. Axial CT images of 357 renal tumor cases were collected. There were 265 (74.2%) cases histologically proven to be malignant, while 34 (9.5%) cases were benign. Radiologists diagnosed 58 (16.3%) cases as angiomyolipoma (AML), based on characteristic appearance, not confirmed histopathologically. For ANN training, the arterial CT phase images were used. A total of 7207 arterial-phase images were collected, then cropped and added to the database with the associated diagnosis. For the test dataset (ANN validation), 38 cases (10 benign, 28 malignant) were chosen by subgroup randomization to correspond to statistical tumor type distribution. The VGG-16 ANN architecture was used in this study. RESULTS Trained ANN correctly classified 23 out of 28 malignant tumors and 8 out of 10 benign tumors. Accuracy was 81.6% (95% confidence interval, 65.7-92.3%), sensitivity was 82.1% (63.1-93.9%), specificity was 80.0% (44.4-97.5%), and F1 score was 86.8% (74.7-94.5%). CONCLUSIONS The created ANN achieved promising accuracy in differentiating benign vs malignant renal tumors.

Forewarned Is Forearmed: Machine Learning Algorithms for the Prediction of Catheter-Induced Coronary and Aortic Injuries.

Catheter-induced dissections (CID) of coronary arteries and/or the aorta are among the most dangerous complications of percutaneous coronary procedures, yet the data on their risk factors are anecdotal. Logistic regression and five more advanced machine learning techniques were applied to determine the most significant predictors of dissection. Model performance comparison and feature importance ranking were evaluated. We identified 124 cases of CID in electronic databases containing 84,223 records of diagnostic and interventional coronary procedures from the years 2000-2022. Based on the f1-score, Extreme Gradient Boosting (XGBoost) was found to have the optimal balance between positive predictive value (precision) and sensitivity (recall). As by the XGBoost, the strongest predictors were the use of a guiding catheter (angioplasty), small/stenotic ostium, radial access, hypertension, acute myocardial infarction, prior angioplasty, female gender, chronic renal failure, atypical coronary origin, and chronic obstructive pulmonary disease. Risk prediction can be bolstered with machine learning algorithms and provide valuable clinical decision support. Based on the proposed model, a profile of 'a perfect dissection candidate' can be defined. In patients with 'a clustering' of dissection predictors, a less aggressive catheter and/or modification of the access site should be considered.

Catheter-induced coronary artery and aortic dissections. A study of the mechanisms, risk factors, and propagation causes.

BACKGROUND: Only the incidence, management, and prognosis of catheter-induced coronary artery and aortic dissections have been systematically studied until now. We sought to evaluate their mechanisms, risk factors, and propagation causes. METHODS: Electronic databases containing 76,104 procedures and complication registries from 2000-2020 were searched and relevant cineangiographic studies adjudicated. RESULTS: Ninety-six dissections were identified. The overall incidence was 0.126%, and 0.021% for aortic injuries. The in-hospital mortality rate was 4.2%, and 6.25% for aortic dissections. Compared to the non-complicated population, patients with dissection were more often female (48% vs. 34%, p = 0.004), with a higher prevalence of comorbidities such as hypertension (56% vs. 25%, p < 0.001) or chronic kidney disease (10% vs. 4%, p = 0.002). They more frequently presented with acute myocardial infarction (72% vs. 43%, p < 0.001), underwent percutaneous coronary intervention (85% vs. 39%, p < 0.001), and were examined with a radial approach (77% vs. 65%, p = 0.011). The most prevalent predisposing factor was small ostium diameter and/or atheroma. Deep intubation for support, catheter malalignment, and vessel prodding were the most frequent precipitating factors. Of the three dissection mechanisms, 'wedged contrast injection' was the commonest (the exclusive mechanism of aortic dissections). The propagation rate was 30.2% and led to doubling of coronary occlusions and aortic extensions. The most frequent progression triggers were repeat injections and unchanged catheter. In 94% of cases, dissections were inflicted by high-volume operators, with ≥ 5-year experience in 84% of procedures. The annual dissection rate increased over a 21-year timespan. CONCLUSIONS: Catheter-induced dissection rarely came unheralded and typically occurred during urgent interventions performed in high-risk patients by experienced operators.